---
title: "CUET_SYNTHETICA@DravidianLangTech 2026: Multi"
description: "This shared-task paper addresses the binary classification of Tamil YouTube comments as Abusive or… 13 references, cited by 1 paper here."
authors: ["Miftahul Jannat Rishta", "Sumaiya Zaman", "Shiti Chowdhury", "Hasan Murad"]
year: 2026
venue: "Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages"
source: "https://aclanthology.org/2026.dravidianlangtech-1.29/"
doi: "10.18653/v1/2026.dravidianlangtech-1.29"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/rishta-2026-cuet-synthetica-dravidianlangtech-2026-multi/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/rishta-2026-cuet-synthetica-dravidianlangtech-2026-multi/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/11/) / CUET\_SYNTHETICA@DravidianLangTech 2026: Multi Architecture Transformer Ensemble for Detecting Abusive Tamil Text Targeting Women

# CUET\_SYNTHETICA@DravidianLangTech 2026: Multi Architecture Transformer Ensemble for Detecting Abusive Tamil Text Targeting Women

**Authors:** [Miftahul Jannat Rishta](https://md.research.elamur.ai/free/en/authors/miftahul-jannat-rishta/), [Sumaiya Zaman](https://md.research.elamur.ai/free/en/authors/sumaiya-zaman/), [Shiti Chowdhury](https://md.research.elamur.ai/free/en/authors/shiti-chowdhury/) and [Hasan Murad](https://md.research.elamur.ai/free/en/authors/hasan-murad/) · **Year:** 2026 · **Venue:** [Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages](https://md.research.elamur.ai/free/en/venues/dravidianlangtech/) ([Workshop on Speech, Vision, and Language Technologies for Dravidian Languages 2026](https://md.research.elamur.ai/free/en/venues/dravidianlangtech/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.dravidianlangtech-1.29/) · [PDF](https://aclanthology.org/2026.dravidianlangtech-1.29.pdf) · [DOI 10.18653/v1/2026.dravidianlangtech-1.29](https://doi.org/10.18653/v1/2026.dravidianlangtech-1.29) · [OpenAlex](https://openalex.org/W7166885716)

**Citations:** 0 citations (OpenAlex, 2026-10-14) · [1](https://md.research.elamur.ai/free/en/papers/rishta-2026-cuet-synthetica-dravidianlangtech-2026-multi/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
This shared-task paper addresses the binary classification of Tamil YouTube comments as Abusive or Non-Abusive, focusing on abuse aimed at women. The setting is the DravidianLangTech@ACL 2026 task. Tamil is hard for this because of its rich morphology, frequent mixing with English, and hostility that is often implicit or sarcastic. The authors normalized the labels, removed duplicate training items, and cleaned the text through a multi-step pipeline. They then fine-tuned three transformers: MuRIL and IndicBERT v2 for Indic and Tamil coverage, and XLM-RoBERTa-Large for broader multilingual signal. Each model was trained four times with different learning rates, using class-weighted cross-entropy and AdamW. The softmax outputs of all twelve runs are averaged (soft voting) to produce the final label. The ensemble reached a macro F1 of 0.8086 on the official test set and placed 4th. It beat every single model and every ensemble built from one architecture alone, though the margins were small. The work suggests that mixing language-specialized and multilingual encoders helps for low-resource abuse detection, while also showing that implicit and sarcastic abuse remains difficult.

## Key points
- **Data:** 3,652 training and 913 test comments, with near-balanced classes. After removing 401 duplicates, 3,251 training samples remained, split 80/20 into 2,600 for training and 651 for validation.
- **Models:** MuRIL (17 Indian languages plus transliterations), IndicBERT v2 (24 languages) and XLM-RoBERTa-Large (100 languages), each run with 4 learning rates for 12 models in total. XLM-R-Large needed much lower rates (2e-6 to 5e-6).
- **Results (macro F1):**
  - Full 12-model ensemble: 0.8086.
  - Best single model, IndicBERT_2: 0.8075.
  - Per-architecture 4-model ensembles: about 0.795 to 0.799.
- **Errors:** The ensemble produced 108 false negatives versus 78 false positives. Missed cases were mostly implicit, culturally coded, sarcastic, or code-mixed abuse.
- **Limitations:** Ensemble weights were equal rather than tuned, with no labeled dev set and no hyperparameter sweeps. Training ran on a single Kaggle Tesla T4 GPU. Code is available at github.com/Mif-taa/Abusive-Tamil-Text.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [Chittagong University of Engineering & Technology](https://md.research.elamur.ai/free/en/universities/chittagong-university-of-engineering-technology/) — University, Chittagong, Bangladesh · 25 works · 29 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🤝 Social Sciences](https://md.research.elamur.ai/free/en/shelves/social-sciences/) · 1,772 works · [🤝 Communication](https://md.research.elamur.ai/free/en/shelves/social-sciences/communication/) · 97 works

## Cite

Cite as: Miftahul Jannat Rishta, Sumaiya Zaman, Shiti Chowdhury and Hasan Murad. 2026. CUET\_SYNTHETICA@DravidianLangTech 2026: Multi Architecture Transformer Ensemble for Detecting Abusive Tamil Text Targeting Women. Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{rishta2026cuet,
  title = {{CUET\_SYNTHETICA@DravidianLangTech 2026: Multi Architecture Transformer Ensemble for Detecting Abusive Tamil Text Targeting Women}},
  author = {Miftahul Jannat Rishta, Sumaiya Zaman, Shiti Chowdhury and Hasan Murad},
  booktitle = {Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages},
  year = {2026},
  doi = {10.18653/v1/2026.dravidianlangtech-1.29},
  url = {https://doi.org/10.18653/v1/2026.dravidianlangtech-1.29}
}
```

- [← previous CUET_InferX@DravidianLangTech 2026: Shared Task on Dialect Based Speech Recognition and Classification in Tamil](https://md.research.elamur.ai/free/en/papers/semon-2026-cuet-inferx-dravidianlangtech-2026-shared/)
- [next → CUET_SYNTHETICA@DravidianLangTech 2026: Multilingual Transformer Based Hope Speech Detection for Coarse and Fine-Grained Classification in Tulu](https://md.research.elamur.ai/free/en/papers/zaman-2026-cuet-synthetica-dravidianlangtech-2026-multilingual/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/11/)

## References

13 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Bharathi Raja Chakravarthi, Ruba Priyadharshini, Vigneshwaran Muralidaran, Navya Jose, Shardul Suryawanshi, Elizabeth Sherly, and John P. McCrae. 2022. Dravidiancodemix: Sentiment analysis and offensive language identification dataset for dravidian languages in code-mixed text. Language Resources an… [find](https://scholar.google.com/scholar?q=Bharathi+Raja+Chakravarthi%2C+Ruba+Priyadharshini%2C+Vigneshwaran+Muralidaran%2C+Navya+Jose%2C+Shardul+Suryawanshi%2C+Elizabeth+Sherly%2C+and+John+P%2E+McCrae%2E+2022%2E+Dravidiancodemix%3A+Sentiment+analysis+and+offensi)
2. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Unsupervised cross-lingual representation learning at scale. In Proceedings of the 58th Annual Meeting of the Association fo… [find](https://scholar.google.com/scholar?q=Alexis+Conneau%2C+Kartikay+Khandelwal%2C+Naman+Goyal%2C+Vishrav+Chaudhary%2C+Guillaume+Wenzek%2C+Francisco+Guzm%C3%A1n%2C+Edouard+Grave%2C+Myle+Ott%2C+Luke+Zettlemoyer%2C+and+Veselin+Stoyanov%2E+2020%2E+Unsupervised+cross%2Dlingu)
3. Sumanth Doddapaneni, Rahul Aralikatte, Gowtham Ramesh, Shreya Goyal, Mitesh M. Khapra, Anoop Kunchukuttan, and Pratyush Kumar. 2023. [Towards leaving no indic language behind: Building monolingual corpora, benchmark and models for indic languages](https://md.research.elamur.ai/free/en/papers/doddapaneni-2023-leaving-no-indic-language-behind/). In Proceedings of the 61st Annual Meeting of the Asso…
4. Bhuvaneswari Sivagnanam, Kathiravan Pannerselvam, Jananayagan V, Charmathi Rajkumar, Ramesh Kannan R, Ratnavel Rajalakshmi, Shunmuga Priya Muthusamy Chinnan, Saranya Rajiakodi, and Bharathi Raja Chakravarthi. 2026. From Comments to Harm: A Findings Report on Abusive Tamil Text Targeting Women on Soc… [find](https://scholar.google.com/scholar?q=Bhuvaneswari+Sivagnanam%2C+Kathiravan+Pannerselvam%2C+Jananayagan+V%2C+Charmathi+Rajkumar%2C+Ramesh+Kannan+R%2C+Ratnavel+Rajalakshmi%2C+Shunmuga+Priya+Muthusamy+Chinnan%2C+Saranya+Rajiakodi%2C+and+Bharathi+Raja+Chakr)
5. Sreelakshmi, B. Premjith, Bharathi Raja Chakravarthi, and K. P. Soman. 2024. Detection of hate speech and offensive language codemix text in dravidian languages using cost-sensitive learning approach. IEEE Access, 12:20064–20090. [find](https://scholar.google.com/scholar?q=Sreelakshmi%2C+B%2E+Premjith%2C+Bharathi+Raja+Chakravarthi%2C+and+K%2E+P%2E+Soman%2E+2024%2E+Detection+of+hate+speech+and+offensive+language+codemix+text+in+dravidian+languages+using+cost%2Dsensitive+learning+approach%2E)
6. Suman Dowlagar and Radhika Mamidi. 2021. OFFLangOne@DravidianLangTech-EACL2021: Transformers with the class balanced loss for offensive language identification in Dravidian code-mixed text. In Proceedings of the First Workshop on Speech and Language Technologies for Dravidian Languages, pages 154–15… [find](https://scholar.google.com/scholar?q=Suman+Dowlagar+and+Radhika+Mamidi%2E+2021%2E+OFFLangOne%40DravidianLangTech%2DEACL2021%3A+Transformers+with+the+class+balanced+loss+for+offensive+language+identification+in+Dravidian+code%2Dmixed+text%2E+In+Proceed)
7. Luxshan Thavarasa, Sivasuthan Sukumar, and Jubeerathan Thevakumar. 2025. In-cepto@DravidianLangTech 2025: Detecting abusive Tamil and Malayalam text targeting women on YouTube. In Proceedings of the Fifth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages. Association for… [find](https://scholar.google.com/scholar?q=Luxshan+Thavarasa%2C+Sivasuthan+Sukumar%2C+and+Jubeerathan+Thevakumar%2E+2025%2E+In%2Dcepto%40DravidianLangTech+2025%3A+Detecting+abusive+Tamil+and+Malayalam+text+targeting+women+on+YouTube%2E+In+Proceedings+of+the+F)
8. Tareque Md Hanif and Md Rashadur Rahman. 2025. CUET_Agile@DravidianLangTech 2025: Fine-tuning transformers for detecting abusive text targeting women from Tamil and Malayalam texts. In Proceedings of the Fifth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages. Association… [find](https://scholar.google.com/scholar?q=Tareque+Md+Hanif+and+Md+Rashadur+Rahman%2E+2025%2E+CUET%5FAgile%40DravidianLangTech+2025%3A+Fine%2Dtuning+transformers+for+detecting+abusive+text+targeting+women+from+Tamil+and+Malayalam+texts%2E+In+Proceedings+of+)
9. Konthala Yasaswini, Karthik Puranik, Adeep Hande, Ruba Priyadharshini, Sajeetha Thavareesan, and Bharathi Raja Chakravarthi. 2021. IIITT@DravidianLangTech-EACL2021: Transfer learning for offensive language detection in Dravidian languages. In Proceedings of the First Workshop on Speech and Language… [find](https://scholar.google.com/scholar?q=Konthala+Yasaswini%2C+Karthik+Puranik%2C+Adeep+Hande%2C+Ruba+Priyadharshini%2C+Sajeetha+Thavareesan%2C+and+Bharathi+Raja+Chakravarthi%2E+2021%2E+IIITT%40DravidianLangTech%2DEACL2021%3A+Transfer+learning+for+offensive+lan)
10. Simran Khanuja, Diksha Bansal, Sarvesh Mehtani, Savya Khosla, Atreyee Dey, Balaji Gopalan, Dilip Kumar Margam, Pooja Aggarwal, Rajiv Teja Nagipogu, Shachi Dave, et al. 2021. [Muril: Multilingual representations for indian languages](https://md.research.elamur.ai/free/en/papers/khanuja-2021-muril-multilingual-representations-indian-languages/). arXiv preprint arXiv:2103.10730.
11. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. [Roberta: A robustly optimized bert pretraining approach](https://md.research.elamur.ai/free/en/papers/liu-2019-roberta-robustly-optimized-bert-pretraining/). In arXiv preprint arXiv:1907.11692.
12. B Premjith, Bharathi Raja Chakravarthi, Prasanna Kumar Kumaresan, Saranya Rajiakodi, Sai Karnati, Sai Mangamuru, and Chandu Janakiram. 2024. Findings of the shared task on hate and offensive language detection in telugu codemixed text (hold-telugu)@ dravidianlangtech 2024. In Proceedings of the Four… [find](https://scholar.google.com/scholar?q=B+Premjith%2C+Bharathi+Raja+Chakravarthi%2C+Prasanna+Kumar+Kumaresan%2C+Saranya+Rajiakodi%2C+Sai+Karnati%2C+Sai+Mangamuru%2C+and+Chandu+Janakiram%2E+2024%2E+Findings+of+the+shared+task+on+hate+and+offensive+language+)
13. Md Mizanur Rahman, Srijita Dhar, Md Mehedi Hasan, and Hasan Murad. 2025. MSM_CUET@DravidianLangTech 2025: XLM-BERT and MuRIL based transformer models for detection of abusive Tamil and Malayalam text targeting women on social media. In Proceedings of the Fifth Workshop on Speech, Vision, and Languag… [find](https://scholar.google.com/scholar?q=Md+Mizanur+Rahman%2C+Srijita+Dhar%2C+Md+Mehedi+Hasan%2C+and+Hasan+Murad%2E+2025%2E+MSM%5FCUET%40DravidianLangTech+2025%3A+XLM%2DBERT+and+MuRIL+based+transformer+models+for+detection+of+abusive+Tamil+and+Malayalam+text+)

*Updated 31 October 2026*